The Life & Mind Seminar Network

Sensitivity to pre-conditions vs. sensitivity to repercussions of self

Posted in Seminars by matthewegbert on December 3, 2009

On Nov. 27, 2009, Tom Froese chaired an interesting Life & Mind seminar in which he proposed two different views of autopoiesis. The first, labelled `Ashbian’, is a horizontal view of autopoiesis in which all of the relevant features (the essential variable(s), and the mechanisms that prevent their leaving their region of viability) are are part of the same, low-level domain of description. No system-level (emergent) properties are necessary to describe such a system. This kind of system maintains certain component conditions which may improve the odds of it as a whole continuing to exist. In other words, they only indirectly maintain system viability through maintaining certain pre-conditions of the system’s existence.

The alternative idea, labelled `Maturanian’ or `Kantian’ – sees the system maintaining its organization not indirectly through the maintenance of component processes or pre-conditions, but a rather more direct maintenance of the organization as a whole. This idea of `maintenance of organization’ is poorly defined or understood. How can a system maintain in general its organization? The notion of homeostasis of a condition, like temperature, is easy to grasp, but maintenance of an organization? How could this be accomplished in a manner more direct than the Ashbian autopoiesis described above that only indirectly maintains its organization through homeostasis of a condition it requires to exist? Is it possible to do anything more than simply maintain the pre-conditions?

This question is one of the motivations behind my current research in understanding how an autopoietic system can be sensitive to its own emergent (or system-as-a-whole) viability. I think there are straightforward mechanisms through which a system can be sensitive to emergent viability. I shall try to describe one such mechanism below. The main underlying idea is the notion of a system being sensitive not only to pre-conditions for its existence (e.g. the presence of resources), but also to effects or repercussions of the system’s existence.

Imagine an system that is Ashbian-autopoietic in its maintenance of a high-availability of a resource it requires to build its components (let’s call it sugar). Imagine now that the resource becomes toxic or alternatively a non-metabolizable analogue like saccharin is present in the environment that confuses this sensitivity. This particular system will not adapt to this change as the essential variable is this measurement of sugar/saccharin. Its actions respond to the pre-condition (the concentration of the original resource) rather than to the effects of that condition.

In the same scenario, a `Maturanian’ autopoietic system: i.e. a system that maintains its organization, would be capable of adapting to the environmental change as its essential variable is total system operational health – its organization. If sugar became toxic, total system operational health would start to fail. The essential variable would move past a critical threshold, and this would cause a system reorganization that would allow for adaptation (e.g. the seeking out of a new resource, or movement to a new location).

I understand that this latter kind of adaptation is observed in E. coli in the following manner [looking for confirmation of this now]. The bacteria are at first attracted to saccharin (a non-metabolizable molecule that confuses certain biological saccharide sensory-systems), but after a duration of
exposure they `realize’ that the environmental feature is not beneficial to the system and cease to be attracted to the sugar analogue [is it no longer attracted to sugar either?]. The bacteria have a sensitivity to an environmental feature that they try to maintain at high concentrations, but they are also sensitive to the repercussions of their sensitivities upon their viability. This latter sensitivity allows them to activate or disactivate or modulate in some more complicated manner their Ashbian environmental-feature-homeostases.

I suggest that biological systems employ both mechanisms. They are sensitive to their own viability through self-monitoring at a variety of scales, including pre-conditions as well as emergent system properties such as the existence and results of macro-molecular structures (e.g. maintenance of high concentration of reactants by membranes) and products of interactions between system components. Our model of an adaptive protocell[1] that responds to the contribution of environmental resources to system health rather than responding directly to environmental phenomena was a first attempt to model this kind of sensitivity (that we are in the process of extending). In that model, the system benefited from being sensitive to the concentration not of resources, but of the result of the presence of those resources: i.e. the concentration of its own constituent parts.

It is sensitive to a simple locally measurable phenomena, but this phenomena is special in that it is indicative of the health of the system as a whole. In our model, the membrane could not be ruptured, but if we imagine that it could, a rupture would cause a drop in concentration of reactants within the membrane, just as might a toxin interfering with an important part of the metabolism of the agent. The point I am trying to make is that there are simple low level phenomena that can indicate general, emergent system-health and that can accordingly be used to influence the system to adapt in general to novel environments. These indicators will never be perfect, but they can be reasonably general.

It seems to me that living systems have a collection of homeostatic mechanisms. Some are sensitive to what I have called pre-conditions; the values being maintained are conditions that in general contribute to system viability (like the sugar/saccharin sensor in the examples above). Others are sensitive to the results of the systems existence (“post-conditions”?). The sensitivity to the pre-conditions is more like what Tom described as Ashbian — it does not involve a sensitivity to emegent-properties of the agent and is therefore not as adaptive (on its own) as a system that also incorporates sensitivities to the repercussions of the system (a.k.a. the results of emergent properties of the system).

My second point is that there exist simple mechanisms for this sensitivity to emergent properties of the system. What Tom described as `Maturanian’ autopoiesis is therefore easily compatible with conventional physical laws and principles.

[1] Egbert, M., Di Paolo, E. A. and Barandiaran, X. (2009)
Chemo-ethology of an Adaptive Protocell: Sensorless sensitivity to
implicit viability conditions in Proceedings of the Tenth European
Conference on Artificial Life, ECAL09, Budapest, September 13-16,
2009, Springer Verlag.

Advertisements

One Response

Subscribe to comments with RSS.

  1. Tom Froese said, on December 4, 2009 at 11:28 am

    Hi Matt,

    Many thanks for your extensive comments! I think you managed to summarize the message of that seminar well, and I like how you immediately think of ways in which the philosophical analysis can be used as an impulse for doing better science.

    Before I add some further thoughts to this debate, let me briefly clarify one important point. In my view at least, I firmly place Maturana’s work within what I have called the “Ashbyan interpretation” of autopoiesis. That is, he strictly denies that there could be any emergent properties having a top-down effect on the parts of the component level (e.g. the doctrine of non-intersecting domains). I contrast this view with the “Kantian interpretation” of autopoiesis, in which I place Varela’s later work and the work we have been doing in the enactive paradigm. The difference comes down whether we allow any sense to the statement “the parts are there because of the whole”, in addition to the standard claim “the whole is there because of the parts”.

    In sum, I see Ashby and Maturana as having a ‘horizontal’ approach, whereby the proper domain of scientific analysis is restricted to the level of components, whereas I see Kant and Varela (and us) as having a ‘hierarchical’ approach, whereby there are several interlocked domains of analysis including the levels of ‘components’ and ‘wholes’.

    Now let’s get to the heart of your post: how do we shift from the Ashbyan model, in which an organization is maintained because an essential variable is maintained, to the Kantian model, where an organization is maintained because of sensitivity to emergent properties (including the viability constraints that are entailed by the organization).

    Finding a principled answer to this question is indeed of great interest. We can note, for example, that the Ashbyan model is actually extremely brittle: if anything in the homeostat actually breaks (let’s say a transistor fries), then it is no longer a homeostat. Also, there is no mechanism by which the homeostat can work against the tendency of a variable to go out of bounds before it induces a pre-defined ‘break’. It’s a purely conservative architecture in that respect.

    Now the Kantian model, if we get it right, has the potential for great sensitivity and open-ended adaptivity. This is because the whole is not just there because of the parts, but because the parts are also there because of the whole. These parts, including the Ashbyan mechanism itself, can be changed and rearranged according to system level properties. As such, we could call this “holistic robustness” (cf. Froese & Ziemke 2009, p. 486).

    And this brings us to what Tom Ziemke and I referred to as the “hard problem of AI”: how do we design Kantian systems that have the appropriate system level behaviors to make use of this open-ended capacity for change? What we need to do is provide the initial conditions whereby a system with emergent properties self-organizes from components in such a way that these properties, in relation to environmental conditions, lead to the emergence of adaptive behavior which rearranges the constituting components of the system appropriately. This is the problem of engineering “second-order emergence”.

    The suggestion that you make is a good one: if the parts are there for the whole, then they will partly reflect the status of that whole, and can therefore be used as indicators for the deployment of adaptive processes. I look forward to seeing you develop your model of this idea so that we can continue this discussion in more concrete terms.

    Cheers,
    Tom

    Froese, T. and Ziemke, T. (2009), “Enactive Artificial Intelligence: Investigating the systemic organization of life and mind”, Journal of Artificial Intelligence, 173(3-4), pp. 466-500 (PDF)


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: